Sectional girder bridge digital modeling and intelligent analysis method

Abstract

The invention provides a segmental girder bridge digital modeling and intelligent analysis method which comprises the following steps: S1, constructing a bridge three-dimensional model through a BIM parametric modeling tool, defining construction stage parameters, and extracting geometric parameters, material attributes, load distribution and boundary conditions of a bridge; s2, discretizing a bridge into a plurality of space nodes along a central axis, generating space sequence data, adjusting node density through a user interaction interface, and inputting node information into a one-dimensional space convolutional network according to a space sequence; s3, physical constraints are constructed in combination with a structural mechanical differential equation, and internal force and deflection distribution are output through an intelligent calculation module; and S4, dynamically associating a calculation result with the BIM model through an IFC standard interface, driving model parameter updating, and forming a'modeling-analysis-optimization 'closed-loop process.

Description

Technical Field

[0001] The present invention relates to the field of intersection of digital modeling technology and artificial intelligence, and in particular to a digital modeling and intelligent analysis method for segmental beam bridges. Background Art

[0002] As an important structural form in modern bridge engineering, the design and construction of segmental beam bridges must balance structural safety and economy. Traditionally, bridge modeling relies on general-purpose BIM software for three-dimensional visual design, but the degree of parameterization is insufficient, making it difficult to dynamically adjust model properties according to the construction phase. Structural analysis typically uses finite element software for numerical simulation, which requires manual meshing and repeated iterative calculations. This is inefficient and has limited adaptability to complex boundary conditions. Furthermore, existing technologies often use independent BIM models and mechanical analysis data, requiring manual synchronization and updating during design optimization. This results in delayed information transmission and makes it difficult to achieve a fully digitalized closed-loop process.

[0003] With the increasing complexity of bridge spans and structural forms, the above methods have exposed the following technical bottlenecks: First, the traditional discretized node processing method cannot efficiently capture the continuity of the axial mechanical behavior of bridges, especially in scenarios such as prestressed reinforcement arrangement and variable cross-section design. The grid dependence leads to an imbalance between calculation accuracy and efficiency; Second, the existing intelligent algorithms in bridge analysis mostly focus on end-to-end data-driven predictions, lack the integration of prior knowledge of structural mechanics differential equations, and are prone to physically unreasonable calculation results; Third, the two-way dynamic correlation mechanism between BIM models and mechanical analysis results has not yet been perfected, and there is still a gap between design parameter adjustment and structural performance verification, which restricts the integrated process of "design-analysis-optimization".

[0004] Therefore, a digital processing and intelligent analysis method for segmental beam bridges is urgently needed to overcome the limitations of traditional tools in modeling efficiency, calculation accuracy and process collaboration. Summary of the Invention

[0005] The purpose of the present invention is to solve the technical problems in the above background and propose a digital modeling and intelligent analysis method for segmental beam bridges, comprising the following steps:

[0006] S1. Build a three-dimensional bridge model using BIM parametric modeling tools, define construction phase parameters, and extract the bridge's geometric parameters, material properties, load distribution, and boundary conditions;

[0007] S2. Discretize the bridge into several spatial nodes along the central axis to generate spatial sequence data. Adjust the node density through the user interface and input the node information into a one-dimensional spatial convolutional network in spatial order.

[0008] S3, combining structural mechanics differential equations to build physical constraints, and outputting internal force and deflection distribution through intelligent computing modules;

[0009] S4. Dynamically associate the calculation results with the BIM model through the IFC standard interface, drive the update of model parameters, and form a "modeling-analysis-optimization" closed-loop process.

[0010] In the preferred solution, the geometric parameters extracted in step S1 include cross-sectional dimensions, span arrangement, and prestressed tendon positions; material properties include elastic modulus, bulk density, and creep coefficient; load distribution includes dead load, live load, and temporary construction load; and boundary conditions include support type, constraint direction, and stiffness value.

[0011] In a preferred solution, step S3 further includes the following steps:

[0012] S31, fitting the displacement field based on the preset basis function library, and calculating the cross-sectional internal force through derivatives;

[0013] S32, using a one-dimensional spatial convolutional network to perform multi-layer convolution calculations to capture the mechanical relationship between adjacent nodes;

[0014] S33. Joint pre-trained boundary network dynamically constrains support conditions.

[0015] In a preferred solution, the intelligent computing module in step S3 includes the following steps:

[0016] S301, normalizing the coordinates of the bridge centerline to generate a training data set of randomly selected points across the entire bridge;

[0017] S302, pre-training boundary network: using support reaction as output, constructing a loss function based on the force balance equation and support stiffness constraints, and independently training the boundary condition prediction model;

[0018] S303, jointly training the one-dimensional spatial convolutional network and the boundary network;

[0019] S304. Output the bending moment, shear force and deflection distribution and compare them with the finite element results.

[0020] In a preferred solution, the basis function library includes preset polynomial functions, trigonometric functions and beam vibration modal functions, and the basis function coefficients are dynamically adjusted through training;

[0021] The convolution kernel of the one-dimensional spatial convolutional network is designed as a learnable differential operator, and each convolution operation corresponds to the discretization calculation of the mechanical differential equation;

[0022] In the preferred solution, the loss function adopts dimension normalization processing, and the displacement, internal force and load are standardized respectively.

[0023] In the preferred solution, the loss function includes:

[0024] Loss of load balance:

[0025]

[0026] Among them, EI is the bending stiffness of the beam, which is composed of the product of the material elastic modulus E and the section inertia moment I, u(x i ) is the bridge in normalized coordinate x i The predicted deflection value at is calculated by fitting the displacement field function of the basis function library. is the displacement function u(x) at x i The fourth-order derivative at , corresponding to the rate of change of the curvature of the beam, q(x i ) is the normalized coordinate x i The distributed load at each location includes external loads such as dead load and live load. N is the total number of random sampling points, which is used to calculate the average error over the entire bridge.

[0027] Loss of internal force balance:

[0028]

[0029] Among them, M(x i ) is the normalized coordinate x i The predicted value of the section bending moment at is directly output by the network or calculated through shear force integration. is the displacement function u(x) at x i The second derivative at corresponds to the rate of change of the curvature of the beam;

[0030] Boundary condition loss:

[0031] L b =Σ k (u(x k )-u target ) 2 +(M(x k )-M target ) 2 ;

[0032] Among them, x k Normalized coordinates of the boundary point, u(xk) is the boundary point x k Predicted deflection value at u target is the boundary point x k The theoretical deflection value at is set according to the support type, M(x k ) is the boundary point x k Predicted bending moment at M target is the boundary point x k Theoretical bending moment value at .

[0033] In the preferred solution, the input of the pre-trained boundary network is the predicted displacement and bending moment at the support, and the output is the support reaction. The loss function is:

[0034] L boundary =∑ j (R j -K j ·u(x j )) 2 ;

[0035] Among them, R j is the predicted value of the support reaction at the j-th support, K j is the stiffness coefficient of the j-th support, which is set according to the actual support type, u(x j ) is the predicted displacement value at the j-th support position xj, which is calculated by the displacement field function.

[0036] In a preferred embodiment, the physical constraints include the equilibrium equations And boundary conditions:

[0037] Fixed end:

[0038] u(x)=0,

[0039] Where u(x) is the displacement at the support position x, is the first derivative of displacement, indicating the cross-sectional rotation angle;

[0040] Simply supported end:

[0041] u(x)=0,M(x)=0;

[0042] Where M(x) is the bending moment at the support position x;

[0043] Free end:

[0044] M(x)=0, V(x)=0;

[0045] Where V(x) is the shear force at the free end position x, which is calculated by the following formula:

[0046]

[0047] In a preferred embodiment, step S4 includes the following steps:

[0048] S41. Map intelligent analysis results to BIM component properties, including cross-sectional size adjustment and prestressed tendon repositioning;

[0049] S42. Automatically trigger model parameter optimization based on deflection exceeding limit results and generate design change suggestions;

[0050] S43. Update the BIM model database through the IFC standard interface and generate visual deformation and internal force cloud maps.

[0051] The beneficial effects of the present invention are:

[0052] 1) Use BIM parametric modeling tools to conveniently extract various aspects of the original data of the bridge, realize dynamic parameter adjustment, improve design efficiency, and reduce parameter update lag.

[0053] 2) Discretize the bridge, combine intelligent computing with dynamic association of BIM models, break down information silos, and promote the integration process.

[0054] 3) Adopt intelligent computing modules, integrate structural mechanics knowledge, efficiently capture the continuity of mechanical behavior, avoid unreasonable results, and ensure calculation accuracy through comparative verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is the BIM model of a segmental beam bridge.

[0056] Figure 2 It is a single-box, single-chamber asymmetric section of a segmental beam bridge.

[0057] Figure 3 It is the digital interface for Revit bridge analysis.

[0058] Figure 4 It is the flow chart for inputting parameters of bridge engineering.

[0059] Figure 5 This is the interface diagram for setting the characteristic values of prestressed steel tendons.

[0060] Figure 6 It is a structural diagram of the temporal convolutional neural network.

[0061] Figure 7 This is a diagram of the intelligent calculation model of segmental beam bridge based on TCN.

[0062] Figure 8 It is a schematic diagram of mechanical quantity fitting based on the basis function library.

[0063] Figure 9 It is a schematic diagram of the pre-trained boundary network.

[0064] Figure 10 This is the MIDAS beam bridge test sample model and calculation results diagram.

[0065] Figure 11 This is a graph of deep learning model training results. DETAILED DESCRIPTION

[0066] Example 1

[0067] Secondary development of BIM parametric modeling for soil segmental beam bridges:

[0068] 1. Automatic identification and extraction method of model parameters:

[0069] Segmental beam bridges are generally composed of dozens to hundreds of concrete box beam segments and piers. Their structural system is relatively clear, and in the parametric modeling process, the segmental box beam general family is usually used to create the box beam entity ( Figure 1 ). Therefore, writing relevant recognition algorithms based on family characteristics can effectively improve efficiency.

[0070] The recognition algorithm uses C# language and first uses the filter class provided by RevitAPI

[0071] The FilteredElementCollector obtains all family instances of the FamilyInstance type in the model space, and then uses the pre-set instance name attribute, such as "single box single room family (symmetric)" or "single box double room family (symmetric)", to filter out different types of instances. For different instance individuals of the same type, the LookupParameter function is used to query their parameters. Finally, all query reading results are summarized and output to form a cross-section parameter summary table for the segment beam bridge BIM model (see Table 1 for some examples of recognition results), as well as the corresponding parameter meanings ( Figure 1 and Figure 2 ).

[0072] Table 1 Identification Results of Some Section Parameters of Concrete Segmental Beam Bridges

[0073]

[0074] 2. Analyze the development process of digital interactive interfaces:

[0075] Bridge calculation and analysis also requires the configuration of engineering parameters such as material properties, prestressed steel, load conditions, and construction phases. To achieve one-stop analysis of segmental beam bridges based on BIM models, this section redevelops a digital interactive interface within the Revit platform based on the WPF (Windows Presentation Foundation) framework.

[0076] WPF is a Windows-based user interface framework launched by Microsoft. Compared with the previous generation WinForms, WPF has an excellent binding mechanism and a modern interface. Using the MVVM (Model-View-ViewModel) architecture can effectively improve development efficiency and achieve better display effects in Revit. The Revit bridge analysis digital interface developed in this paper mainly includes the operation panel interface ( Figure 3 ) and the corresponding subwindows for each button. The operation panel interface includes three functional modules: tendon settings, pre-settings and construction stage settings, and data processing. The tendon settings module includes buttons for tendon characteristics, tendon shape, and tendon prestressing. The pre-settings and construction stage settings module includes buttons for pre-settings, construction stage definition, and construction stage settings. Node import and file export buttons are also provided for the model reading and subsequent data file organization functions mentioned above.

[0077] During the modeling phase, users can select different function buttons in sequence to enter relevant parameters: define the type of prestressed tendons used in the tendon properties; set the specific tendon parameters and shapes used in the project in the tendon shape settings; set the relevant parameters for tensioning prestressing in the tendon prestressing settings; predefine the bridge structure group, tendon group, load group, and boundary group in the presetting, and set relevant parameters including secondary loads and temperature loads; finally, define all construction phases and durations of bridge construction, as well as the specific construction steps of the construction phase, including the activation and passivation settings of the structure group, load group activation, prestressing settings, and boundary condition activation and passivation.

[0078] For the specific sub-interfaces that implement each function, the operator's convenience is improved by setting a drop-down list, setting default options based on the specific characteristics of segmental beam bridges, displaying the entered data through a data table, and allowing addition, editing, and deletion operations. Figure 5 Taking the prestressed steel tendon characteristic value setting interface as an example, the sub-interface layout and function implementation are demonstrated.

[0079] 3. Organization and transmission

[0080] Automatically extracted and user-entered model data is stored in a backend database. However, directly outputting it to the intelligent computing module without proper structure and organization can easily lead to data redundancy and confusion. Therefore, this section uses the open-source MCT format as the basic data transmission format. Furthermore, the data conversion process can also perform missing and format checks on the input data.

[0081] MCT files are command stream input files for the commercial bridge analysis software MIDAS CIVIL. They are equivalent to MIDAS CIVIL model files (MCB files) and can be converted through import and export. MCT files are primarily composed of different parameter modules, which correspond one-to-one to the interactive parameter input method of MIDAS. The basic format of each command module is:

[0082] *Command;Command comment

[0083] ; Parameter name 1, parameter name 2...; parameter comment 1

[0084] ; Parameter name 1, parameter name 2, parameter name 3...; parameter comment 2

[0085] ;……

[0086] Parameter 1, parameter 2...

[0087] Parameter 1, parameter 2, parameter 3...

[0088] Code is written to import command modules such as node definition, element definition, structure group, boundary group, load group, material and time-dependent properties, tendon group, tendon properties, tendon arrangement, boundary constraints, load definition, and construction phase from the back-end data table according to the above format. All functions are integrated into the MCT file output button mentioned above to realize the automatic generation of data files. The following is the construction phase part of the generated MCT file sample.

[0089] *STAGE;Define Construction Stage

[0090] NAME=Two spans hoisted, 60, YES, NO, NO, 5

[0091] AELEM = Span 1,90, Span 2,90, Temporary support 1,90, Temporary support 2,90, Temporary support 3,90

[0092] ABNDR = Temporary support 1, DEFORMED, Temporary support 2, DEFORMED, Temporary support 3, DEFORMED

[0093] ALOAD = Self-weight, FIRST, Tendon-first span, FIRST, Tendon-second span, FIRST

[0094] NAME=Middle Pier Cast-in-place,30,YES,NO,NO,5

[0095] AELEM=cast-in-place section, 90

[0096] ALOAD=top beam, FIRST, bottom beam, FIRST

[0097] NAME=System Transformation,7,YES,NO,NO,5

[0098] AELEM=Fixed support 1,90,Fixed support 2,90,Fixed support 3,90

[0099] DELEM = Temporary support 1,100, Temporary support 2,100, Temporary support 3,100

[0100] ABNDR = Fixed support 1, DEFORMED, Fixed support 2, DEFORMED, Fixed support 3, DEFORMED

[0101] DBNDR = Temporary support 1, Temporary support 2, Temporary support 3

[0102] NAME=Second phase dead load,30,YES,NO,NO,5

[0103] ALOAD=secondary load, FIRST

[0104] NAME=Ten Years of Change,3650,YES,NO,NO,5

[0105] This completes the development of a pre-processing interface for BIM-based parametric modeling and analysis of concrete segmental beams. Users can easily modify parameters to quickly update the BIM model, with the program automatically identifying and extracting those parameters. Furthermore, engineering parameters required for analysis can be defined and organized on the same platform, avoiding the duplication of work and asynchronous updates that can arise from model migration. This seamlessly links pre-processing and subsequent calculation procedures, laying the foundation for one-stop design services for segmental beam bridges.

[0106] The above solution has the following advantages:

[0107] (1) A parametric input and extraction program for concrete segmental beams was developed based on universal families and Revit API, which can realize the rapid digital update and automatic identification of segmental beam bridge information.

[0108] (2) An engineering parameter user interaction interface for the BIM platform was developed based on the WPF framework, which facilitates users to input the data and information required for bridge analysis and avoids problems such as duplication of work and asynchronous updates that may be caused by platform migration.

[0109] Example 2

[0110] Bridge intelligent calculation method:

[0111] To achieve one-stop analysis of segmental beam bridges, this section proposes a theory-driven intelligent calculation method for beam bridges. This method includes a data organization method for beam bridges and a core intelligent calculation model. This method incorporates structural mechanics knowledge to ensure the theoretical correctness of the calculation results.

[0112] 1. Data organization

[0113] Feature data is the foundation of intelligent technology, and its organization determines the architecture of subsequent processing models. As a common bridge structure, beam bridges have the following characteristics:

[0114] (1) The topological structure of a beam bridge in space presents a single linear shape. It has a central axis in three-dimensional space. In the spatial flow coordinate system, the three-dimensional coordinates can be represented by a single coordinate. At the same time, its basic parameters can be established along the central axis. That is, the mapping between its properties such as material and cross-section can be established through a unique coordinate value.

[0115] (2) Beam bridge nodes exhibit both sequential and continuous spatial properties. On the one hand, beam bridges exhibit a certain degree of spatial directionality; on the other hand, the relative positional relationships between different spans and nodes remain unchanged. Combined with the first characteristic, beam bridge data can be input not only as a whole but also sequentially, depending on the requirements.

[0116] (3) The stress pattern of a beam bridge conforms to the theoretical derivation of structural mechanics. A beam bridge is formed by connecting different spans through multiple supports. The negative moment zone is generally connected to form a whole through post-casting and prestressed tendons. Its structural form and constraint method conform to the basic assumptions of structural mechanics, and therefore can be embedded in structural mechanics equations as prior knowledge.

[0117] (4) The forces acting on beam bridges have local effects. Under local loads, the nearby beam segments are more affected, while the distal beam segments are less affected. Since the spatial span of a beam bridge is relatively large relative to the bridge cross-section, the overall stiffness is small. The calculated values of the cross-sectional internal forces and deflections are generally related to the loads and performance of adjacent bridge segments. From the traditional finite element method, this is reflected in the fact that the stiffness matrix has smaller values away from the diagonal.

[0118] Based on the above characteristics, after conducting a feasibility analysis of different data structures such as vectors, sequences, and graphs, it was finally determined that sequences would be used as the pre-processing format for deep learning networks: after subdividing the bridge into several nodes, the node information is input sequentially in spatial order. The sequence data format can, to a certain extent, avoid the problem of uncertain input dimensions in the vector format (caused by reasons such as the length of the beam bridge and the number of divided segments), and effectively improve the distinction between different nodes through the sequential transmission of information. At the same time, the memory effect of the sequence model can be used to better perceive the information of adjacent nodes. In addition, compared to the more expressive graph data structure, the sequence format is less complex and simpler, thus fully leveraging the efficient computing capabilities of the intelligent model. Therefore, a corresponding intelligent model will be developed based on the sequence data format.

[0119] 2. Driven Temporal Convolutional Network

[0120] Based on the analysis of beam bridge data characteristics, a one-dimensional temporal convolutional neural network (TCN, such as Figure 6 As shown in the figure, the core deep learning model is established. On this basis, a pre-trained boundary neural network is added to control the boundary conditions and jointly trained with the calculation results of the convolutional network. The overall framework is as follows Figure 7 shown.

[0121] Different from the discrete grid in the finite element method, this model adopts the idea of continuous function for fitting calculation. Since the deflection and bending moment curves in general beam bridges are highly characteristic, a function library is constructed by pre-setting basis functions (including polynomial functions and trigonometric functions), and the coefficients of different basis functions are trained through deep learning, such as Figure 8 As shown in the figure. By using a continuous function, a mapping of loads and cross-sectional parameters to the bridge's axial coordinates can be established, freeing the loss function calculation from being limited by the fineness of the mesh. Based on this continuous function, the coordinates of the entire bridge are normalized, and during training and calculation, random points are taken from the entire bridge as data as needed, and these random points are serialized. In TCN, the initial data undergoes multi-layer convolution calculations, with adjacent data points calculated sequentially.

[0122] The loss function defines the degree of difference between the neural network's calculated results and the true value, and is used to perform backpropagation and updates of the neural network. The loss function in this model consists of several components: load balance loss (L1), moment internal force balance loss (L2), and support force balance loss (L3). The balance losses of L1 and L2 are derived from structural mechanics. Since the derivatives of the basis functions are easily obtained, based on the basic principle of the Euler beam, the cross-sectional internal force distribution and the node reaction force distribution can be calculated by taking multiple derivatives of the deflection distribution function.

[0123] The pretrained boundary network pretrains neural networks N1 and N2 using boundary conditions, ensuring that the output consistently meets the boundary constraints. This reduces the constraints on boundary deflections and rotations in the TCN, improving computational efficiency. The results of the pretrained network characterize the relative positional relationship between any point on the bridge and its adjacent supports. In practical implementation, the TCN is jointly trained with the pretrained networks N1 and N2, and the backpropagation gradient is calculated based on the loss function.

[0124] 3. Calculation Method Verification

[0125] The intelligent calculation method for beam bridges proposed above was tested on a 3×30m continuous beam bridge and compared with the analysis results of MIDAS software. The bridge section adopts a concrete single-box single-chamber section, such as Figure 10 shown.

[0126] The bridge is modeled in BIM and then input into the TCN model. The intelligent model calculation results are as follows: Figure 11 and as shown in Table 2. During the training process, the model converged quickly, the calculated bending moments had little error with the MIDAS calculation results, and the shape of the bending moment curves had a high degree of coincidence: the maximum positive bending moment error of the beam bridge was 2.46%, the maximum negative bending moment error at the support was 2.71%, and the maximum deflection error at midspan was 5.63%, all within the acceptable error range. This indicates that the training converged well, the calculation results were relatively accurate, and the intelligent calculation method can meet the needs of bridge analysis.

[0127] Table 2 Comparison between Results of Deep Learning Model and FEA

[0128]

[0129] The intelligent calculation method for beam bridges can be well connected with the BIM-based parametric modeling method by quickly parsing the exported MCT data files. In the future, by decomposing and calculating the various working conditions in the bridge construction and completion stages and then superimposing and combining them, a complete analysis of segmental beam bridges can be achieved, thus forming a one-stop digital and intelligent design and analysis service for segmental beam bridges.

[0130] Numerical tests show that the calculation results of this method are consistent with those of finite element calculations, and the accuracy can meet the requirements of bridge analysis.

[0131] The intelligent calculation method can be well connected with the MCT files obtained by BIM parametric modeling. In the future, in-depth research on the applicability of intelligent models will be conducted for various bridge working conditions, and ultimately a one-stop digital and intelligent design and analysis service will be formed.

[0132] Example 3

[0133] A digital modeling and intelligent analysis method for segmental beam bridges includes the following steps:

[0134] S1. Build a three-dimensional bridge model using BIM parametric modeling tools, define construction phase parameters, and extract the bridge's geometric parameters, material properties, load distribution, and boundary conditions;

[0135] S2. Discretize the bridge into several spatial nodes along the central axis to generate spatial sequence data. Adjust the node density through the user interface and input the node information into a one-dimensional spatial convolutional network in spatial order.

[0136] S3, combining structural mechanics differential equations to build physical constraints, and outputting internal force and deflection distribution through intelligent computing modules;

[0137] S4. Dynamically associate the calculation results with the BIM model through the IFC standard interface, drive the update of model parameters, and form a "modeling-analysis-optimization" closed-loop process.

[0138] Preferably, the geometric parameters extracted in step S1 include cross-sectional dimensions, span arrangement, and prestressed tendon positions; material properties include elastic modulus, bulk density, and creep coefficient; load distribution includes dead load, live load, and temporary construction load; and boundary conditions include support type, constraint direction, and stiffness value.

[0139] Preferably, step S3 further includes the following steps:

[0140] S31, fitting the displacement field based on the preset basis function library, and calculating the cross-sectional internal force through derivatives;

[0141] S32, using a one-dimensional spatial convolutional network to perform multi-layer convolution calculations to capture the mechanical relationship between adjacent nodes;

[0142] S33. Joint pre-trained boundary network dynamically constrains support conditions.

[0143] Preferably, the intelligent computing module in step S3 includes the following steps:

[0144] S301, normalizing the coordinates of the bridge centerline to generate a training data set of randomly selected points across the entire bridge;

[0145] S302, pre-training boundary network: using support reaction as output, constructing a loss function based on the force balance equation and support stiffness constraints, and independently training the boundary condition prediction model;

[0146] S303, jointly training the one-dimensional spatial convolutional network and the boundary network;

[0147] S304. Output the bending moment, shear force and deflection distribution and compare them with the finite element results.

[0148] Preferably, the basis function library includes preset polynomial functions, trigonometric functions and beam vibration modal functions, and the basis function coefficients are dynamically adjusted through training;

[0149] The convolution kernel of the one-dimensional spatial convolutional network is designed as a learnable differential operator, and each convolution operation corresponds to the discretization calculation of the mechanical differential equation;

[0150] Preferably, the loss function adopts dimensional normalization processing, and the displacement, internal force and load are standardized respectively.

[0151] Preferably, the loss function includes:

[0152] Loss of load balance:

[0153]

[0154] Among them, EI is the bending stiffness of the beam, which is composed of the product of the material elastic modulus E and the section inertia moment I, u(x i ) is the bridge in normalized coordinate x i The predicted deflection value at is calculated by fitting the displacement field function of the basis function library. is the displacement function u(x) at x i The fourth-order derivative at , corresponding to the rate of change of the curvature of the beam, q(x i ) is the normalized coordinate x i The distributed load at each location includes external loads such as dead load and live load. N is the total number of random sampling points, which is used to calculate the average error over the entire bridge.

[0155] Loss of internal force balance:

[0156]

[0157] Among them, M(x i ) is the normalized coordinate x i The predicted value of the section bending moment at is directly output by the network or calculated through shear force integration. is the displacement function u(x) at x i The second derivative at corresponds to the rate of change of the curvature of the beam;

[0158] Boundary condition loss:

[0159] L b =Σ k (u(x k )-u target ) 2 +(M(x k )-M target )2 ;

[0160] Among them, x k Normalized coordinates of the boundary point, u(xk) is the boundary point x k Predicted deflection value at u target is the boundary point x k The theoretical deflection value at is set according to the support type, M(x k ) is the boundary point x k Predicted bending moment at M target is the boundary point x k Theoretical bending moment value at .

[0161] Preferably, the input of the pre-trained boundary network is the predicted displacement and bending moment at the support, and the output is the support reaction force. The loss function is:

[0162] L boundary =∑ j (R j -K j ·u(x j )) 2 ;

[0163] Among them, R j is the predicted value of the support reaction at the j-th support, K j is the stiffness coefficient of the j-th support, which is set according to the actual support type, u(x j ) is the predicted displacement value at the j-th support position xj, which is calculated by the displacement field function.

[0164] Preferably, the physical constraints include equilibrium equations And boundary conditions:

[0165] Fixed end:

[0166] u(x)=0,

[0167] Where u(x) is the displacement at the support position x, is the first derivative of displacement, indicating the cross-sectional rotation angle;

[0168] Simply supported end:

[0169] u(x)=0, M(x)=0;

[0170] Where M(x) is the bending moment at the support position x;

[0171] Free end:

[0172] M(x)=0,V(x)=0;

[0173] Where V(x) is the shear force at the free end position x, which is calculated by the following formula:

[0174]

[0175] Preferably, step S4 includes the following steps:

[0176] S41. Map intelligent analysis results to BIM component properties, including cross-sectional size adjustment and prestressed tendon repositioning;

[0177] S42. Automatically trigger model parameter optimization based on deflection exceeding limit results and generate design change suggestions;

[0178] S43. Update the BIM model database through the IFC standard interface and generate visual deformation and internal force cloud maps.

[0179] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A digital modeling and intelligent analysis method for segmental beam bridges, characterized by: The following steps are involved: S1. Build a three-dimensional bridge model using BIM parametric modeling tools, define construction phase parameters, and extract the bridge's geometric parameters, material properties, load distribution, and boundary conditions; S2. Discretize the bridge into several spatial nodes along the central axis to generate spatial sequence data. Adjust the node density through the user interface and input the node information into a one-dimensional spatial convolutional network in spatial order. S3, combining structural mechanics differential equations to build physical constraints, and outputting internal force and deflection distribution through intelligent computing modules; S4. Dynamically associate the calculation results with the BIM model through the IFC standard interface, drive the update of model parameters, and form a "modeling-analysis-optimization" closed-loop process.

2. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 1 is characterized by: The geometric parameters extracted in step S1 include cross-sectional dimensions, span arrangement, and prestressed tendon location; material properties include elastic modulus, bulk density, and creep coefficient; load distribution includes dead load, live load, and temporary construction load; and boundary conditions include support type, constraint direction, and stiffness value.

3. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 1 is characterized by: Step S3 further includes the following steps: S31, fitting the displacement field based on the preset basis function library, and calculating the cross-sectional internal force through derivatives; S32, using a one-dimensional spatial convolutional network to perform multi-layer convolution calculations to capture the mechanical relationship between adjacent nodes; S33. Joint pre-trained boundary network dynamically constrains support conditions.

4. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 1 is characterized by: The intelligent calculation module in step S3 includes the following steps: S301, normalizing the coordinates of the bridge centerline to generate a training data set of randomly selected points across the entire bridge; S302, pre-training boundary network: using support reaction as output, constructing a loss function based on the force balance equation and support stiffness constraints, and independently training the boundary condition prediction model; S303, jointly training the one-dimensional spatial convolutional network and the boundary network; S304. Output the bending moment, shear force and deflection distribution and compare them with the finite element results.

5. According to claim 3, a digital modeling and intelligent analysis method for segmental beam bridges is characterized by: The function library includes preset polynomial functions, trigonometric functions and beam vibration modal functions, and dynamically adjusts the basis function coefficients through training; The convolution kernel of the one-dimensional spatial convolutional network is designed as a learnable differential operator, and each layer of convolution operation corresponds to the discretization calculation of the mechanical differential equation.

6. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 4 is characterized by: The loss function adopts dimension normalization to standardize the displacement, internal force and load respectively.

7. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 4 is characterized by: The loss functions include: Loss of load balance: Among them, EI is the bending stiffness of the beam, which is composed of the product of the material elastic modulus E and the section inertia moment I, u(x i ) is the bridge in normalized coordinate x i The predicted deflection value at is calculated by fitting the displacement field function of the basis function library. is the displacement function u(x) at x i The fourth-order derivative at , corresponding to the rate of change of the curvature of the beam, q(x i ) is the normalized coordinate x i The distributed load at each location includes external loads such as dead load and live load. N is the total number of random sampling points, which is used to calculate the average error over the entire bridge. Loss of internal force balance: Among them, M(x i ) is the normalized coordinate x i The predicted value of the section bending moment at is directly output by the network or calculated through shear force integration. is the displacement function u(x) at x i The second derivative at corresponds to the rate of change of the curvature of the beam; Boundary condition loss: L b =Σ k (u(x k )-u target ) 2 +(M(x k )-M target ) 2 ; Among them, x k Normalized coordinates of the boundary point, u(xk) is the boundary point x k Predicted deflection value at u target is the boundary point x k The theoretical deflection value at is set according to the support type, M(x k ) is the boundary point x k Predicted bending moment at M target is the boundary point x k Theoretical bending moment value at .

8. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 3 or 4, characterized in that: The input of the pre-trained boundary network is the predicted displacement and bending moment at the support, and the output is the support reaction. The loss function is: L boundary =∑ j (R j -K j ·u(x j )) 2 ; Among them, R j is the predicted value of the support reaction at the j-th support, K j is the stiffness coefficient of the j-th support, which is set according to the actual support type, x(x j ) is the predicted displacement value at the j-th support position xj, which is calculated by the displacement field function.

9. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 1 is characterized by: Physical constraints include equilibrium equations And boundary conditions: Fixed end: Where u(x) is the displacement at the support position x, is the first derivative of displacement, indicating the cross-sectional rotation angle; Simply supported end: u(x)=0,M(x)=0; Where M(x) is the bending moment at the support position x; Free end: M(x)=0,V(x)=0; Where V(x) is the shear force at the free end position x, which is calculated by the following formula:

10. The digital modeling and intelligent analysis method for segmental beam bridges according to claim 1 is characterized by: Step S4 includes the following steps: S41. Map intelligent analysis results to BIM component properties, including cross-sectional size adjustment and prestressed tendon repositioning; S42. Automatically trigger model parameter optimization based on deflection exceeding limit results and generate design change suggestions; S43. Update the BIM model database through the IFC standard interface and generate visual deformation and internal force cloud maps.

Images